recall

Recall is a lightweight open-source plugin that gives Claude Code persistent memory across sessions, entirely on your machine. Every new Claude Code session starts cold — Recall maintains a local log of your sessions and condenses it into a resume-ready summary using a classical Python summarizer (TF-IDF + TextRank). No API key, no external model, no data ever leaves your machine. It costs nothing beyond your Claude Code subscription and saves tokens by reducing repetitive context re-explanation. Recall works by writing two files into your project's .recall/ folder: history.md (the append-only session log) and context.md (the compact summary overwritten each session). The summarizer runs locally — no LLM calls, no token spend. Features include automatic session logging, auto-save on session end, and a /recall:save command for manual checkpoints. The integration is minimal: Claude Code's stop/session-start hooks trigger capture and context loading. For privacy-conscious developers, Recall is a clear win — your code, paths, and any secrets are never sent to an API. Most competing memory tools pipe context to a cloud model, whereas Recall's privacy guarantee is absolute. It's complementary to Claude Code's built-in memory (CLAUDE.md, /continue, or context compaction), filling the gap with an automatic, deterministic record of what each session did. Compared to alternatives like Memanto or Reyn, Recall is narrower — it only supports Claude Code and uses extractive summarization rather than LLM-based memory. That makes it simpler, cheaper (free), and more private, but less capable of understanding complex relationships in context. It's ideal for solo developers or small teams running Claude Code locally who want continuity without the complexity or cost of a memory server.